[
https://issues.apache.org/jira/browse/ARTEMIS-5379?focusedWorklogId=964717&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-964717
]
ASF GitHub Bot logged work on ARTEMIS-5379:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 03/Apr/25 16:29
Start Date: 03/Apr/25 16:29
Worklog Time Spent: 10m
Work Description: clebertsuconic commented on PR #5595:
URL:
https://github.com/apache/activemq-artemis/pull/5595#issuecomment-2776347011
What I'm doing in this PR is the following:
I added two methods on transaction. Delay and DelayDone
When Delay is called I increase a counter.
when delayDone is called I set the counter back to zero, and check if
there's any pending task.
Also.. when commit (or prepare) is called, if there are pending delays
instead of recording the commit, I set a task to be done after delayDone is
called.
Beyone that everything is testing.. I added two tests: one UnitTest using
Mock, and one SoakTest based on a series of EAP tests that will complete in
hours (mine takes 6 seconds).
Issue Time Tracking
-------------------
Worklog Id: (was: 964717)
Time Spent: 20m (was: 10m)
> Journal commit record recorded before Page Writer is finished leading to
> possible message loss on a crash
> ---------------------------------------------------------------------------------------------------------
>
> Key: ARTEMIS-5379
> URL: https://issues.apache.org/jira/browse/ARTEMIS-5379
> Project: ActiveMQ Artemis
> Issue Type: Bug
> Reporter: Clebert Suconic
> Priority: Blocker
> Labels: pull-request-available
> Time Spent: 20m
> Remaining Estimate: 0h
>
> Due to ARTEMIS-5305, the journal commit record could reach the journal before
> the page writes.
> The clients were always waiting correctly for the entire portion of the data
> being stored. However if an uncertainty happened (the failure happened before
> the page write was finished) you could have a situation where part of the TX
> is recorded and part is not.
> That could lead to DUPLICATE Cache stored and not the data.
> Or other portion of the TX stored but not the send.
> The fix is to defer the write of the journal record until the page write
> operation is finished.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
For further information, visit: https://activemq.apache.org/contact
